Polygon

Seven Parquet tables with 269 documented columns in total: blocks (22), transactions (55), calls (42), events (45), transfers (24), dex_trades (54) and miner_rewards (27). Covering the most recent month to yesterday, refreshed daily.

On Transaction_Hash, which appears in every table except blocks and miner_rewards; those two join on Block_Number. Within a transaction, Call_Index orders the call frames, Log_Index orders the logs, Transfer_Index orders the transfers and Trade_Index orders the swaps. A log points back to the frame that emitted it through Call_Index.

Two things. The Polymarket prediction tables — prediction trades, prediction settlements and prediction managements — are not here; they are sold as the separate Polymarket dataset. Balance updates are also excluded. If you need either, ask and we will quote them alongside this one.

The dedicated Polymarket tables are not, and neither is the matic/dex_trades/polymarket prefix. But dex_trades is a decode of all swap activity on Polygon, and Polymarket order fills are on-chain swaps on Polygon, so rows whose Trade_Dex_ProtocolFamily is Polymarket do appear in it. Filter that column out if you want Polymarket excluded entirely — it is a large share of recent Polygon swap volume, so the difference is not marginal.

The layout was set before the MATIC to POL rebrand and we did not move it, because moving it would break every existing manifest and download script. Paths are matic/blocks, matic/transactions and so on. Nothing about the data is stale — only the prefix name.

No. Polygon's Bor consensus produces no uncle blocks, so Block_UnclesCount is 0 on every block, Block_UncleHash is the empty-list hash, and Reward_Uncle, Uncle_Block_Number and Uncle_Index in miner_rewards are empty or zero throughout. The columns exist because the schema is shared with chains that do have uncles; they are kept rather than dropped so a query written against one EVM chain runs unchanged here.

Amounts are decimal-adjusted decimal strings, carrying exactly as many places as the token has decimals — multiply by ten to the power of the decimals for the raw on-chain integer. Casting to float loses precision on large values. Booleans are the strings true and false, not 0 and 1, in every column.

Less than you might expect, and the dataset says so rather than filling the gap. Only six currencies have an hourly price series on Polygon — POL, WPOL, WETH, USDT, DAI and the bridged USDC.e — which prices roughly a fifth of transfer rows; the rest is mostly airdrop and spam tokens. DEX trades do much better, because each swap is priced from its safer side and most swaps have a stablecoin or POL on one of them. A USD value of 0 means either a zero-value row or no price; the column cannot tell the two apart.

Yes, throughout, and the export applies no filter of its own. TransactionStatus_Success carries the transaction's outcome on every table that has it, and dex_trades additionally carries Trade_Success, because a single swap can fail inside a transaction that succeeded. Spam tokens, zero-value rows and lookalike contracts are all kept too — the columns let you drop what you do not want.

Filter Call_To in calls and Log_SmartContract in events to the contract, lower case with 0x. For token flow, filter Transfer_Currency_SmartContract in transfers. Join on contract addresses rather than symbols or names — both repeat, and lookalike tokens copy them deliberately.

One line in DuckDB: SELECT * FROM read_parquet('transfers/*.parquet') LIMIT 10. Pandas works too: pandas.read_parquet(path). The sample files open without credentials.

As flat Parquet named by block range, 50 blocks per file and sorted ascending, one prefix per table, with a JSON manifest listing every file and its sha256. Signed HTTPS links are emailed once the files are prepared, or on request we deliver into your own S3, GCS or R2 bucket. Refreshed daily with T+1 latency, so a purchase made today includes everything to yesterday.

Yes. Every table links to a public sample file with real records, and the sample bucket holds full Parquet files you can load directly. No email address required.